zircon/system/ulib/blobfs/host.cc

// Copyright 2017 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include <fcntl.h>
#include <inttypes.h>
#include <lib/cksum.h>
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include <memory>
#include <new>
#include <optional>
#include <utility>

#include <blobfs/format.h>
#include <blobfs/host.h>
#include <blobfs/host/fsck.h>
#include <digest/digest.h>
#include <digest/merkle-tree.h>
#include <fbl/algorithm.h>
#include <fbl/array.h>
#include <fbl/auto_call.h>
#include <fbl/macros.h>
#include <fs-host/common.h>
#include <fs/trace.h>
#include <fs/transaction/block_transaction.h>
#include <safemath/checked_math.h>

#include "compression/compressor.h"
#include "compression/zstd.h"

using digest::Digest;
using digest::MerkleTreeCreator;
using digest::MerkleTreeVerifier;

constexpr uint32_t kExtentCount = 5;

namespace blobfs {
namespace {

using HostCompressor = ZSTDCompressor;
const auto HostDecompressor = ZSTDDecompress;
constexpr uint32_t kBlobFlagCompressed = kBlobFlagZSTDCompressed;

zx_status_t ReadBlockOffset(int fd, uint64_t bno, off_t offset, void* data) {
  off_t off = offset + bno * kBlobfsBlockSize;
  if (pread(fd, data, kBlobfsBlockSize, off) != kBlobfsBlockSize) {
    FS_TRACE_ERROR("blobfs: cannot read block %" PRIu64 "\n", bno);
    return ZX_ERR_IO;
  }
  return ZX_OK;
}

zx_status_t WriteBlockOffset(int fd, uint64_t bno, off_t offset, const void* data) {
  off_t off = offset + bno * kBlobfsBlockSize;
  if (pwrite(fd, data, kBlobfsBlockSize, off) != kBlobfsBlockSize) {
    FS_TRACE_ERROR("blobfs: cannot write block %" PRIu64 "\n", bno);
    return ZX_ERR_IO;
  }
  return ZX_OK;
}

// From a buffer, create a merkle tree.
//
// Given a mapped blob at |blob_data| of length |length|, compute the
// Merkle digest and the output merkle tree as a uint8_t array.
zx_status_t buffer_create_merkle(const FileMapping& mapping, MerkleInfo* out_info) {
  zx_status_t status;
  std::unique_ptr<uint8_t []> merkle_tree;
  size_t merkle_size;
  if ((status = MerkleTreeCreator::Create(mapping.data(), mapping.length(), &merkle_tree,
                                          &merkle_size, &out_info->digest)) != ZX_OK) {
    return status;
  }
  out_info->merkle.reset(merkle_tree.release(), merkle_size);
  out_info->length = mapping.length();
  return ZX_OK;
}

zx_status_t buffer_compress(const FileMapping& mapping, MerkleInfo* out_info) {
  size_t max = HostCompressor::BufferMax(mapping.length());
  out_info->compressed_data.reset(new uint8_t[max]);
  out_info->compressed = false;

  if (mapping.length() < kCompressionMinBytesSaved) {
    return ZX_OK;
  }

  zx_status_t status;
  std::unique_ptr<HostCompressor> compressor;
  if ((status = HostCompressor::Create(mapping.length(), out_info->compressed_data.get(), max,
                                       &compressor)) != ZX_OK) {
    FS_TRACE_ERROR("Failed to initialize blobfs compressor: %d\n", status);
    return status;
  }

  if ((status = compressor->Update(mapping.data(), mapping.length())) != ZX_OK) {
    FS_TRACE_ERROR("Failed to update blobfs compressor: %d\n", status);
    return status;
  }

  if ((status = compressor->End()) != ZX_OK) {
    FS_TRACE_ERROR("Failed to complete blobfs compressor: %d\n", status);
    return status;
  }

  if (mapping.length() > compressor->Size() + kCompressionMinBytesSaved) {
    out_info->compressed_length = compressor->Size();
    out_info->compressed = true;
  }

  return ZX_OK;
}

// Given a buffer (and pre-computed merkle tree), add the buffer as a
// blob in Blobfs.
zx_status_t blobfs_add_mapped_blob_with_merkle(Blobfs* bs, FileSizeRecorder* size_recorder,
                                               const FileMapping& mapping, const MerkleInfo& info) {
  ZX_ASSERT(mapping.length() == info.length);
  const void* data;

  if (info.compressed) {
    data = info.compressed_data.get();
  } else {
    data = mapping.data();
  }

  // After we've pre-calculated all necessary information, actually add the
  // blob to the filesystem itself.
  static std::mutex add_blob_mutex_;
  std::lock_guard<std::mutex> lock(add_blob_mutex_);
  std::unique_ptr<InodeBlock> inode_block;
  zx_status_t status;
  if ((status = bs->NewBlob(info.digest, &inode_block)) != ZX_OK) {
    FS_TRACE_ERROR("error: Failed to allocate a new blob\n");
    return status;
  }
  if (inode_block == nullptr) {
    FS_TRACE_ERROR("error: No nodes available on blobfs image\n");
    return ZX_ERR_NO_RESOURCES;
  }

  Inode* inode = inode_block->GetInode();
  inode->blob_size = mapping.length();
  uint64_t block_count = MerkleTreeBlocks(*inode) + info.GetDataBlocks();
  if (block_count > std::numeric_limits<uint32_t>::max()) {
    FS_TRACE_ERROR("error: Block count too large\n");
    return ZX_ERR_NOT_SUPPORTED;
  }
  inode->block_count = static_cast<uint32_t>(block_count);
  inode->header.flags |= kBlobFlagAllocated | (info.compressed ? kBlobFlagCompressed : 0);

  // TODO(smklein): Currently, host-side tools can only generate single-extent
  // blobs. This should be fixed.
  if (inode->block_count > kBlockCountMax) {
    FS_TRACE_ERROR("error: Blobs larger than %lu blocks not yet implemented\n", kBlockCountMax);
    return ZX_ERR_NOT_SUPPORTED;
  }

  size_t start_block = 0;
  if ((status = bs->AllocateBlocks(inode->block_count, &start_block)) != ZX_OK) {
    FS_TRACE_ERROR("error: No blocks available\n");
    return status;
  }

  // TODO(smklein): This is hardcoded alongside the check against "kBlockCountMax" above.
  inode->extents[0].SetStart(start_block);
  inode->extents[0].SetLength(static_cast<BlockCountType>(inode->block_count));
  inode->extent_count = 1;

  if (size_recorder) {
    size_recorder->AppendSizeInformation(info.digest.ToString().c_str(),
                                         kBlobfsBlockSize * inode->block_count);
  }

  if ((status = bs->WriteData(inode, info.merkle.data(), data)) != ZX_OK) {
    return status;
  }

  if ((status = bs->WriteBitmap(inode->block_count, inode->extents[0].Start())) != ZX_OK) {
    return status;
  } else if ((status = bs->WriteNode(std::move(inode_block))) != ZX_OK) {
    return status;
  } else if ((status = bs->WriteInfo()) != ZX_OK) {
    return status;
  }

  return ZX_OK;
}

// Returns ZX_OK and copies blobfs info_block_t, which is a block worth of data containing
// superblock, into |out_info_block| if the block read from fd belongs to blobfs.
zx_status_t blobfs_load_info_block(const fbl::unique_fd& fd, info_block_t* out_info_block,
                                   off_t start = 0, std::optional<off_t> end = std::nullopt) {
  info_block_t info_block;

  if (ReadBlockOffset(fd.get(), 0, start, (void*)info_block.block) < 0) {
    return ZX_ERR_IO;
  }
  uint64_t blocks;
  zx_status_t status;
  if ((status = GetBlockCount(fd.get(), &blocks)) != ZX_OK) {
    FS_TRACE_ERROR("blobfs: cannot find end of underlying device\n");
    return status;
  }

  if (end &&
      ((blocks * kBlobfsBlockSize) < safemath::checked_cast<uint64_t>(end.value() - start))) {
    FS_TRACE_ERROR("blobfs: Invalid file size\n");
    return ZX_ERR_BAD_STATE;
  } else if ((status = CheckSuperblock(&info_block.info, blocks)) != ZX_OK) {
    FS_TRACE_ERROR("blobfs: Info check failed\n");
    return status;
  }

  memcpy(out_info_block, &info_block, sizeof(*out_info_block));

  return ZX_OK;
}

zx_status_t get_superblock(const fbl::unique_fd& fd, off_t start, std::optional<off_t> end,
                           Superblock* info) {
  info_block_t info_block;
  zx_status_t status;

  if ((status = blobfs_load_info_block(fd, &info_block, start, end)) != ZX_OK) {
    return status;
  }

  memcpy(info, &info_block.info, sizeof(info_block.info));
  return ZX_OK;
}

}  // namespace

zx_status_t ReadBlock(int fd, uint64_t bno, void* data) {
  off_t off = bno * kBlobfsBlockSize;
  if (pread(fd, data, kBlobfsBlockSize, off) != kBlobfsBlockSize) {
    FS_TRACE_ERROR("blobfs: cannot read block %" PRIu64 "\n", bno);
    return ZX_ERR_IO;
  }
  return ZX_OK;
}

zx_status_t WriteBlock(int fd, uint64_t bno, const void* data) {
  off_t off = bno * kBlobfsBlockSize;
  if (pwrite(fd, data, kBlobfsBlockSize, off) != kBlobfsBlockSize) {
    FS_TRACE_ERROR("blobfs: cannot write block %" PRIu64 "\n", bno);
    return ZX_ERR_IO;
  }
  return ZX_OK;
}

zx_status_t GetBlockCount(int fd, uint64_t* out) {
  struct stat s;
  if (fstat(fd, &s) < 0) {
    return ZX_ERR_BAD_STATE;
  }
  *out = s.st_size / kBlobfsBlockSize;
  return ZX_OK;
}

int Mkfs(int fd, uint64_t block_count) {
  Superblock info;
  InitializeSuperblock(block_count, &info);
  zx_status_t status = CheckSuperblock(&info, block_count);
  if (status != ZX_OK) {
    FS_TRACE_ERROR("Failed to initialize superblock: %d\n", status);
    return -1;
  }
  uint64_t block_bitmap_blocks = BlockMapBlocks(info);
  uint64_t node_map_blocks = NodeMapBlocks(info);

  RawBitmap block_bitmap;
  if (block_bitmap.Reset(block_bitmap_blocks * kBlobfsBlockBits)) {
    FS_TRACE_ERROR("Couldn't allocate blobfs block map\n");
    return -1;
  } else if (block_bitmap.Shrink(info.data_block_count)) {
    FS_TRACE_ERROR("Couldn't shrink blobfs block map\n");
    return -1;
  }

  // Reserve first |kStartBlockMinimum| data blocks
  block_bitmap.Set(0, kStartBlockMinimum);

  // All in-memory structures have been created successfully. Dump everything to disk.
  char block[kBlobfsBlockSize] = {};

  // Write the journal to disk.
  // 1) Write the Info block.
  // 2) Write the rest of the journal as zeros.
  fs::JournalInfo* journal_info = reinterpret_cast<fs::JournalInfo*>(block);
  journal_info->magic = fs::kJournalMagic;
  journal_info->checksum =
      crc32(0, reinterpret_cast<const uint8_t*>(block), sizeof(fs::JournalInfo));
  if ((status = WriteBlock(fd, JournalStartBlock(info), block)) != ZX_OK) {
    FS_TRACE_ERROR("Failed to write journal block\n");
    return -1;
  }
  for (uint64_t n = 1; n < JournalBlocks(info); n++) {
    memset(block, 0, sizeof(block));
    status = WriteBlock(fd, JournalStartBlock(info) + n, block);
    if (status != ZX_OK) {
      FS_TRACE_ERROR("Failed to write journal block\n");
      return -1;
    }
  }

  // Write the root block to disk.
  memset(block, 0, sizeof(block));
  memcpy(block, &info, sizeof(info));
  if ((status = WriteBlock(fd, 0, block)) != ZX_OK) {
    FS_TRACE_ERROR("Failed to write root block\n");
    return -1;
  }

  // Write allocation bitmap to disk.
  for (uint64_t n = 0; n < block_bitmap_blocks; n++) {
    void* bmdata = GetRawBitmapData(block_bitmap, n);
    if ((status = WriteBlock(fd, BlockMapStartBlock(info) + n, bmdata)) != ZX_OK) {
      FS_TRACE_ERROR("Failed to write blockmap block %" PRIu64 "\n", n);
      return -1;
    }
  }

  // Write node map to disk.
  for (uint64_t n = 0; n < node_map_blocks; n++) {
    memset(block, 0, sizeof(block));
    if (WriteBlock(fd, NodeMapStartBlock(info) + n, block)) {
      FS_TRACE_ERROR("blobfs: failed writing inode map\n");
      return -1;
    }
  }

  FS_TRACE_DEBUG("BLOBFS: mkfs success\n");
  return 0;
}

zx_status_t UsedDataSize(const fbl::unique_fd& fd, uint64_t* out_size, off_t start,
                         std::optional<off_t> end) {
  Superblock info;
  zx_status_t status;

  if ((status = get_superblock(fd, start, end, &info)) != ZX_OK) {
    return status;
  }

  *out_size = info.alloc_block_count * info.block_size;
  return ZX_OK;
}

zx_status_t UsedInodes(const fbl::unique_fd& fd, uint64_t* out_inodes, off_t start,
                       std::optional<off_t> end) {
  Superblock info;
  zx_status_t status;

  if ((status = get_superblock(fd, start, end, &info)) != ZX_OK) {
    return status;
  }

  *out_inodes = info.alloc_inode_count;
  return ZX_OK;
}

zx_status_t UsedSize(const fbl::unique_fd& fd, uint64_t* out_size, off_t start,
                     std::optional<off_t> end) {
  Superblock info;
  zx_status_t status;

  if ((status = get_superblock(fd, start, end, &info)) != ZX_OK) {
    return status;
  }

  *out_size = (TotalNonDataBlocks(info) + info.alloc_block_count) * info.block_size;
  return ZX_OK;
}

zx_status_t blobfs_create(std::unique_ptr<Blobfs>* out, fbl::unique_fd fd) {
  info_block_t info_block;
  zx_status_t status;

  if ((status = blobfs_load_info_block(fd, &info_block)) != ZX_OK) {
    return status;
  }

  fbl::Array<size_t> extent_lengths(new size_t[kExtentCount], kExtentCount);

  extent_lengths[0] = BlockMapStartBlock(info_block.info) * kBlobfsBlockSize;
  extent_lengths[1] = BlockMapBlocks(info_block.info) * kBlobfsBlockSize;
  extent_lengths[2] = NodeMapBlocks(info_block.info) * kBlobfsBlockSize;
  extent_lengths[3] = JournalBlocks(info_block.info) * kBlobfsBlockSize;
  extent_lengths[4] = DataBlocks(info_block.info) * kBlobfsBlockSize;

  if ((status = Blobfs::Create(std::move(fd), 0, info_block, extent_lengths, out)) != ZX_OK) {
    FS_TRACE_ERROR("blobfs: mount failed; could not create blobfs\n");
    return status;
  }

  return ZX_OK;
}

zx_status_t blobfs_create_sparse(std::unique_ptr<Blobfs>* out, fbl::unique_fd fd, off_t start,
                                 off_t end, const fbl::Vector<size_t>& extent_vector) {
  if (start >= end) {
    FS_TRACE_ERROR("blobfs: Insufficient space allocated\n");
    return ZX_ERR_INVALID_ARGS;
  }
  if (extent_vector.size() != kExtentCount) {
    FS_TRACE_ERROR("blobfs: Incorrect number of extents\n");
    return ZX_ERR_INVALID_ARGS;
  }

  info_block_t info_block;
  zx_status_t status;

  if ((status = blobfs_load_info_block(fd, &info_block, start, end)) != ZX_OK) {
    return status;
  }

  fbl::Array<size_t> extent_lengths(new size_t[kExtentCount], kExtentCount);

  extent_lengths[0] = extent_vector[0];
  extent_lengths[1] = extent_vector[1];
  extent_lengths[2] = extent_vector[2];
  extent_lengths[3] = extent_vector[3];
  extent_lengths[4] = extent_vector[4];

  if ((status = Blobfs::Create(std::move(fd), start, info_block, extent_lengths, out)) != ZX_OK) {
    FS_TRACE_ERROR("blobfs: mount failed; could not create blobfs\n");
    return status;
  }

  return ZX_OK;
}

zx_status_t blobfs_preprocess(int data_fd, bool compress, MerkleInfo* out_info) {
  FileMapping mapping;
  zx_status_t status = mapping.Map(data_fd);
  if (status != ZX_OK) {
    return status;
  }

  if ((status = buffer_create_merkle(mapping, out_info)) != ZX_OK) {
    return status;
  }

  if (compress) {
    status = buffer_compress(mapping, out_info);
  }

  return status;
}

zx_status_t blobfs_add_blob(Blobfs* bs, FileSizeRecorder* size_recorder, int data_fd) {
  FileMapping mapping;
  zx_status_t status = mapping.Map(data_fd);
  if (status != ZX_OK) {
    return status;
  }

  // Calculate the actual Merkle tree.
  MerkleInfo info;
  status = buffer_create_merkle(mapping, &info);
  if (status != ZX_OK) {
    return status;
  }

  return blobfs_add_mapped_blob_with_merkle(bs, size_recorder, mapping, info);
}

zx_status_t blobfs_add_blob_with_merkle(Blobfs* bs, FileSizeRecorder* size_recorder, int data_fd,
                                        const MerkleInfo& info) {
  FileMapping mapping;
  zx_status_t status = mapping.Map(data_fd);
  if (status != ZX_OK) {
    return status;
  }

  return blobfs_add_mapped_blob_with_merkle(bs, size_recorder, mapping, info);
}

zx_status_t blobfs_fsck(fbl::unique_fd fd, off_t start, off_t end,
                        const fbl::Vector<size_t>& extent_lengths) {
  std::unique_ptr<Blobfs> blob;
  zx_status_t status;
  if ((status = blobfs_create_sparse(&blob, std::move(fd), start, end, extent_lengths)) != ZX_OK) {
    return status;
  }
  bool apply_journal = false;
  if ((status = Fsck(std::move(blob), apply_journal)) != ZX_OK) {
    return status;
  }
  return ZX_OK;
}

Blobfs::Blobfs(fbl::unique_fd fd, off_t offset, const info_block_t& info_block,
               const fbl::Array<size_t>& extent_lengths)
    : blockfd_(std::move(fd)), dirty_(false), offset_(offset) {
  ZX_ASSERT(extent_lengths.size() == kExtentCount);
  memcpy(&info_block_, info_block.block, kBlobfsBlockSize);
  cache_.bno = 0;

  block_map_start_block_ = extent_lengths[0] / kBlobfsBlockSize;
  block_map_block_count_ = extent_lengths[1] / kBlobfsBlockSize;
  node_map_start_block_ = block_map_start_block_ + block_map_block_count_;
  node_map_block_count_ = extent_lengths[2] / kBlobfsBlockSize;
  journal_start_block_ = node_map_start_block_ + node_map_block_count_;
  journal_block_count_ = extent_lengths[3] / kBlobfsBlockSize;
  data_start_block_ = journal_start_block_ + journal_block_count_;
  data_block_count_ = extent_lengths[4] / kBlobfsBlockSize;
}

zx_status_t Blobfs::Create(fbl::unique_fd blockfd_, off_t offset, const info_block_t& info_block,
                           const fbl::Array<size_t>& extent_lengths, std::unique_ptr<Blobfs>* out) {
  zx_status_t status = CheckSuperblock(&info_block.info, TotalBlocks(info_block.info));
  if (status < 0) {
    FS_TRACE_ERROR("blobfs: Check info failure\n");
    return status;
  }

  ZX_ASSERT(extent_lengths.size() == kExtentCount);

  for (unsigned i = 0; i < 3; i++) {
    if (extent_lengths[i] % kBlobfsBlockSize) {
      return ZX_ERR_INVALID_ARGS;
    }
  }

  auto fs =
      std::unique_ptr<Blobfs>(new Blobfs(std::move(blockfd_), offset, info_block, extent_lengths));

  if ((status = fs->LoadBitmap()) < 0) {
    FS_TRACE_ERROR("blobfs: Failed to load bitmaps\n");
    return status;
  }

  *out = std::move(fs);
  return ZX_OK;
}

zx_status_t Blobfs::LoadBitmap() {
  zx_status_t status;
  if ((status = block_map_.Reset(block_map_block_count_ * kBlobfsBlockBits)) != ZX_OK) {
    return status;
  } else if ((status = block_map_.Shrink(info_.data_block_count)) != ZX_OK) {
    return status;
  }
  const void* bmstart = block_map_.StorageUnsafe()->GetData();

  for (size_t n = 0; n < block_map_block_count_; n++) {
    void* bmdata = fs::GetBlock(kBlobfsBlockSize, bmstart, n);

    if (n >= node_map_start_block_) {
      memset(bmdata, 0, kBlobfsBlockSize);
    } else if ((status = ReadBlock(block_map_start_block_ + n)) != ZX_OK) {
      return status;
    } else {
      memcpy(bmdata, cache_.blk, kBlobfsBlockSize);
    }
  }
  return ZX_OK;
}

zx_status_t Blobfs::NewBlob(const Digest& digest, std::unique_ptr<InodeBlock>* out) {
  size_t ino = info_.inode_count;

  for (size_t i = 0; i < info_.inode_count; ++i) {
    size_t bno = (i / kBlobfsInodesPerBlock) + node_map_start_block_;

    zx_status_t status;
    if ((status = ReadBlock(bno)) != ZX_OK) {
      return status;
    }

    auto iblk = reinterpret_cast<const Inode*>(cache_.blk);
    auto observed_inode = &iblk[i % kBlobfsInodesPerBlock];
    if (observed_inode->header.IsAllocated() && !observed_inode->header.IsExtentContainer()) {
      if (digest == observed_inode->merkle_root_hash) {
        return ZX_ERR_ALREADY_EXISTS;
      }
    } else if (ino >= info_.inode_count) {
      // If |ino| has not already been set to a valid value, set it to the
      // first free value we find.
      // We still check all the remaining inodes to avoid adding a duplicate blob.
      ino = i;
    }
  }

  if (ino >= info_.inode_count) {
    return ZX_ERR_NO_RESOURCES;
  }

  size_t bno = (ino / kBlobfsInodesPerBlock) + NodeMapStartBlock(info_);
  zx_status_t status;
  if ((status = ReadBlock(bno)) != ZX_OK) {
    return status;
  }

  Inode* inodes = reinterpret_cast<Inode*>(cache_.blk);

  std::unique_ptr<InodeBlock> ino_block(
      new InodeBlock(bno, &inodes[ino % kBlobfsInodesPerBlock], digest));

  dirty_ = true;
  info_.alloc_inode_count++;
  *out = std::move(ino_block);
  return ZX_OK;
}

zx_status_t Blobfs::AllocateBlocks(size_t nblocks, size_t* blkno_out) {
  zx_status_t status;
  if ((status = block_map_.Find(false, 0, block_map_.size(), nblocks, blkno_out)) != ZX_OK) {
    return status;
  } else if ((status = block_map_.Set(*blkno_out, *blkno_out + nblocks)) != ZX_OK) {
    return status;
  }

  info_.alloc_block_count += nblocks;
  return ZX_OK;
}

zx_status_t Blobfs::WriteBitmap(size_t nblocks, size_t start_block) {
  uint64_t block_bitmap_start_block = start_block / kBlobfsBlockBits;
  uint64_t block_bitmap_end_block =
      fbl::round_up(start_block + nblocks, kBlobfsBlockBits) / kBlobfsBlockBits;
  const void* bmstart = block_map_.StorageUnsafe()->GetData();
  for (size_t n = block_bitmap_start_block; n < block_bitmap_end_block; n++) {
    const void* data = fs::GetBlock(kBlobfsBlockSize, bmstart, n);
    uint64_t bno = block_map_start_block_ + n;
    zx_status_t status;
    if ((status = WriteBlock(bno, data)) != ZX_OK) {
      return status;
    }
  }

  return ZX_OK;
}

zx_status_t Blobfs::WriteNode(std::unique_ptr<InodeBlock> ino_block) {
  if (ino_block->GetBno() != cache_.bno) {
    return ZX_ERR_ACCESS_DENIED;
  }

  dirty_ = false;
  return WriteBlock(cache_.bno, cache_.blk);
}

zx_status_t Blobfs::WriteData(Inode* inode, const void* merkle_data, const void* blob_data) {
  const size_t merkle_blocks = MerkleTreeBlocks(*inode);
  const size_t data_blocks = inode->block_count - merkle_blocks;
  for (size_t n = 0; n < merkle_blocks; n++) {
    const void* data = fs::GetBlock(kBlobfsBlockSize, merkle_data, n);
    uint64_t bno = data_start_block_ + inode->extents[0].Start() + n;
    zx_status_t status;
    if ((status = WriteBlock(bno, data)) != ZX_OK) {
      return status;
    }
  }

  for (size_t n = 0; n < data_blocks; n++) {
    const void* data = fs::GetBlock(kBlobfsBlockSize, blob_data, n);

    // If we try to write a block, will it be reaching beyond the end of the
    // mapped file?
    size_t off = n * kBlobfsBlockSize;
    uint8_t last_data[kBlobfsBlockSize];
    if (inode->blob_size < off + kBlobfsBlockSize) {
      // Read the partial block from a block-sized buffer which zero-pads the data.
      memset(last_data, 0, kBlobfsBlockSize);
      memcpy(last_data, data, inode->blob_size - off);
      data = last_data;
    }

    uint64_t bno = data_start_block_ + inode->extents[0].Start() + merkle_blocks + n;
    zx_status_t status;
    if ((status = WriteBlock(bno, data)) != ZX_OK) {
      return status;
    }
  }

  return ZX_OK;
}

zx_status_t Blobfs::WriteInfo() { return WriteBlock(0, info_block_); }

zx_status_t Blobfs::ReadBlock(size_t bno) {
  if (dirty_) {
    return ZX_ERR_ACCESS_DENIED;
  }

  zx_status_t status;
  if ((cache_.bno != bno) &&
      ((status = ReadBlockOffset(blockfd_.get(), bno, offset_, &cache_.blk)) != ZX_OK)) {
    return status;
  }

  cache_.bno = bno;
  return ZX_OK;
}

zx_status_t Blobfs::WriteBlock(size_t bno, const void* data) {
  return WriteBlockOffset(blockfd_.get(), bno, offset_, data);
}

zx_status_t Blobfs::ResetCache() {
  if (dirty_) {
    return ZX_ERR_ACCESS_DENIED;
  }

  if (cache_.bno != 0) {
    memset(cache_.blk, 0, kBlobfsBlockSize);
    cache_.bno = 0;
  }
  return ZX_OK;
}

Inode* Blobfs::GetNode(uint32_t index) {
  size_t bno = node_map_start_block_ + index / kBlobfsInodesPerBlock;

  if (bno >= data_start_block_) {
    // Set cache to 0 so we can return a pointer to an empty inode
    if (ResetCache() != ZX_OK) {
      return nullptr;
    }
  } else if (ReadBlock(bno) < 0) {
    return nullptr;
  }

  auto iblock = reinterpret_cast<Inode*>(cache_.blk);
  return &iblock[index % kBlobfsInodesPerBlock];
}

zx_status_t Blobfs::VerifyBlob(uint32_t node_index) {
  Inode inode = *GetNode(node_index);

  // Determine size for (uncompressed) data buffer.
  uint64_t data_blocks = BlobDataBlocks(inode);
  uint64_t merkle_blocks = MerkleTreeBlocks(inode);
  uint64_t num_blocks = data_blocks + merkle_blocks;
  size_t target_size;
  if (mul_overflow(num_blocks, kBlobfsBlockSize, &target_size)) {
    FS_TRACE_ERROR("Multiplication overflow");
    return ZX_ERR_OUT_OF_RANGE;
  }

  // Create data buffer.
  std::unique_ptr<uint8_t[]> data(new uint8_t[target_size]);
  if (inode.header.flags & kBlobFlagCompressed) {
    // Read in uncompressed merkle blocks.
    for (unsigned i = 0; i < merkle_blocks; i++) {
      ReadBlock(data_start_block_ + inode.extents[0].Start() + i);
      memcpy(data.get() + (i * kBlobfsBlockSize), cache_.blk, kBlobfsBlockSize);
    }

    // Determine size for compressed data buffer.
    size_t compressed_blocks = (inode.block_count - merkle_blocks);
    size_t compressed_size;
    if (mul_overflow(compressed_blocks, kBlobfsBlockSize, &compressed_size)) {
      FS_TRACE_ERROR("Multiplication overflow");
      return ZX_ERR_OUT_OF_RANGE;
    }

    // Create compressed data buffer.
    std::unique_ptr<uint8_t[]> compressed_data(new uint8_t[compressed_size]);

    // Read in all compressed blob data.
    for (unsigned i = 0; i < compressed_blocks; i++) {
      ReadBlock(data_start_block_ + inode.extents[0].Start() + i + merkle_blocks);
      memcpy(compressed_data.get() + (i * kBlobfsBlockSize), cache_.blk, kBlobfsBlockSize);
    }

    // Decompress the compressed data into the target buffer.
    zx_status_t status;
    target_size = inode.blob_size;
    uint8_t* data_ptr = data.get() + (merkle_blocks * kBlobfsBlockSize);
    if ((status = HostDecompressor(data_ptr, &target_size, compressed_data.get(),
                                   &compressed_size)) != ZX_OK) {
      return status;
    }
    if (target_size != inode.blob_size) {
      FS_TRACE_ERROR("Failed to fully decompress blob (%zu of %" PRIu64 " expected)\n", target_size,
                     inode.blob_size);
      return ZX_ERR_IO_DATA_INTEGRITY;
    }
  } else {
    // For uncompressed blobs, read entire blob straight into the data buffer.
    for (unsigned i = 0; i < inode.block_count; i++) {
      ReadBlock(data_start_block_ + inode.extents[0].Start() + i);
      memcpy(data.get() + (i * kBlobfsBlockSize), cache_.blk, kBlobfsBlockSize);
    }
  }

  // Verify the contents of the blob.
  uint8_t* data_ptr = data.get() + (merkle_blocks * kBlobfsBlockSize);
  MerkleTreeVerifier mtv;
  zx_status_t status;
  if ((status = mtv.SetDataLength(inode.blob_size)) != ZX_OK ||
      (status = mtv.SetTree(data.get(), mtv.GetTreeLength(), inode.merkle_root_hash,
                            sizeof(inode.merkle_root_hash))) != ZX_OK ||
      (status = mtv.Verify(data_ptr, inode.blob_size, 0)) != ZX_OK) {
    return status;
  }
  return ZX_OK;
}

}  // namespace blobfs